An Investigation of Phone Upgrades in Remote Community Cellular Networks Kushal Shah Philip Martinez Emre Tepedelenlioglu University of Washington University of the Diliman Facebook [email protected] [email protected] [email protected] Shaddi Hasan Cedric Festin Joshua Blumenstock Facebook University of the Philippines Diliman University of California, Berkeley [email protected] [email protected] [email protected] Josephine Dionisio Kurtis Heimerl University of the Philippines Diliman University of Washington/Facebook [email protected] [email protected] ABSTRACT 1 INTRODUCTION In the last decade, billions of people worldwide have upgraded from For billions of people, the primary technology for accessing the basic feature phones to data-enabled . In most cellular network is a 2G [14]. However, as telecoms cases, people upgrade in areas with 4G coverage (typically cities invest heavily in and 4G infrastructure, it is becoming clear that and large towns), but increasingly, people choose to upgrade in the will be the dominant technology of the future. In areas that only have 2G coverage or no cellular coverage at all. This most dense urban areas, the three network types — 2G, 3G, and 4G counterintuitive behavior – upgrading your phone despite living — already operate simultaneously and users can upgrade as their in an area that does not actively support many of the features of devices and disposable income allows. that new device – is the focus of this work. In rural areas, however, it is less clear if and how this transition We investigate the rates and reasons for 4G upgrades and adop- will occur. Particularly in areas without existing coverage, operators tion in two extremely remote areas in Indonesia and the Philippines. face a conundrum in deciding whether to deploy 2G networks that Our mixed-methods approach combines the quantitative analysis of are cheap and broadly compatible (but may soon be obsolete), or several years of registration logs with the qualitative 3G or 4G networks that are modern and expensive (and where they analysis of multiple interviews in one of these communities. We may never recoup their initial investment). Part of this ambiguity learn that users are rapidly switching from 2G to 4G technology stems from the lack of reliable data on how feature phones and and skipping 3G entirely; the data suggest that these villages will smartphones are used in remote communities, and when and why soon have sufficient 4G phone adoption to justify the investment subscribers typically upgrade their devices. required to upgrade base stations to 4G technology. The interviews The goal of this study is to develop an understanding of mobile suggest people are making these switches primarily to support phone upgrade behavior in rural areas. We focus our analysis on consumption of media such as games, videos, and music. Similarly, two rural communities in the Philippines and Indonesia that first users switch devices because of damage, often leading to down- received cellular network access in 2013 and 2016, respectively. We grades to more resilient feature phones. We also find that, despite develop a quantitative picture of phone upgrade behavior through the general value seen in more modern 4G phones, 2G phones are the analysis of a large volume of longitudinal log data collected more shared and more active on the network. by the cellular network operator. Specifically, we analyze the Type Allocation Codes (TACs) logs, which provide a detailed record of ACM Reference format: the phone models used by each user, over time, in each community. Kushal Shah, Philip Martinez, Emre Tepedelenlioglu, Shaddi Hasan, Cedric Festin, Joshua Blumenstock, Josephine Dionisio, and Kurtis Heimerl. 2017. We complement this quantitative analysis with follow-up qualita- An Investigation of Phone Upgrades in Remote Community Cellular Net- tive field work in the Philippines. Through fifteen semi-structured works. In Proceedings of International Conference on Information & Com- interviews with individuals who recently upgraded their phones, munication Technologies and Development, , , November 2017 we add some texture and insight into the stories behind the data. (ACM ICTD ’17), 12 pages. Our primary findings are that: https://doi.org/10.1145/3001913.xxxxxxx • 4G phones are common (17.9% in Indonesia and 22.5% in Permission to make digital or hard copies of part or all of this work for personal or the Philippines), even though the local network does not classroom use is granted without fee provided that copies are not made or distributed support 4G data, for profit or commercial advantage and that copies bear this notice and the full citation • 4G phone adoption has accelerated dramatically on both on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s). networks (more than 2x year-over-year on both networks), ACM ICTD ’17, November 2017, Lahore, Pakistan • Most phone upgrades are between 2G and 4G (93.62% for © 2017 Copyright held by the owner/author(s). Indonesia and 81.53% for the Philippines), skipping 3G ACM ISBN 978-1-4503-5277-2...$15.00 https://doi.org/10.1145/3001913.xxxxxxx almost entirely, ACM ICTD ’17, November 2017, Lahore, Pakistan K. Shah et al.

• Users of 4G phones were less active on the 2G network 2.3 Entertainment as Development than users of 2G only phones, One of the key findings from this work is that entertainment is • People share their 2G phones nearly 3 times more than driving 4G cellular adoption. This is supported by previous studies their 4G phones, that show the importance of entertainment in promoting invest- • Those upgrading to 4G phones were usually young and ment and the use of ICTs. Smyth et al. [30] showed that users in often handed their old phones to family members, and the slums of Bengaluru build complex bluetooth-based networks • A primary reason people upgraded to a 4G phone was to for media sharing. Raza et al. [25] built a system utilizing enter- be able to better play media. tainment (specifically viral voice modifying) to direct users toa voice job board. Our work supports these results and shows that From these findings, we conclude that telecoms should focus on entertainment is similarly important for promoting rural access. only 2G and 4G networks in rural areas. Our back-of-the-envelope calculations indicate that operators could profitably deploy 4G 2.4 Cellular Records networks (rather than 2G) in these communities within the next two years. Finally, we discuss several mechanisms to accelerate 4G Finally, this study relates to a growing body of research using cel- adoption in rural areas. lular records to understand processes of development. Examples include the use of mobile phone data to measure poverty [2], under- stand the impacts of mobile money [4], and more generally model 2 RELATED WORK human behavior [3, 12]. The vast majority of such work uses call data records (CDRs), which are captures of metadata from phone 2.1 Community Cellular calls and text messages. This paper leverages data from two community cellular networks Our use of Type Allocation Codes (TACs) is somewhat novel. (CCNs) installed in remote parts of the world. CCNs [16] are lo- Two major works have explored this data source. Heimerl et al. [18] cally owned and operated cellular networks. There are a number of investigated smartphone adoption in an Indonesian community these operating in the world now, with Rhizomatica [26] being the cellular network and found that the adoption rate was in line with largest. Two other networks include the Papuacom Network [17] the rest of the country and SMS usage was higher on smartphones. in Indonesia and the Kwiizya network in Zambia [1, 34]. There are We use this as justification that the analysis of community cellular three key things about CCNs: first, they are logically separate from networks could potentially be generalized to non-community net- existing telecoms and thus provide a unique sandbox for experi- works. Ahmad et al, [1] did a similar analysis on a regional scale, ments; second, they are operated by local agents and; third, they gathering TACs from the Pakistani province of the Punjab. From usually operate in rural areas without traditional cellular coverage. that dataset they were able to measure operating system versions and capabilities of the phones. There are two key differences from our work. First, Ahmad’s work is a snapshot of the country and 2.2 Mobile Adoption and Use has no longitudinal components, required for measuring smart- phone adoption. Second, it is a trace from a mostly urban province There is a rich literature on technology adoption which we do not and would not assist in exploring the possibilities in rural cellular attempt to survey here [cf. 27]. Venkatesh et al. [33] develop a installations. framework that helps structure our thinking of mobile adoption in this context, particularly with respect to facilitating conditions, social influence, and performance expectancy. Here, facilitating con- 3 CONTEXT ditions, which indicate “the degree to which an individual believes The contexts we study are rural communities in the Philippines that an organizational and technical infrastructure exists to support and Indonesia. While global adoption of smartphones is tracked use of the system,” are relatively weak in these villages – in both closely by several industry groups, very little information exists on cases there is no 3G or 4G coverage in the surveyed communities. rural adoption – this is a key focus and motivation for this study. Regarding social influence, “the degree to which an individual per- Broadly, GSMA Intelligence [19] reports that the overall rate of ceives that important others believe he or she should use the new smartphone adoption in the Philippines in 2014 was 28%, and a system,” is perhaps more limited than in urban areas, although we similar estimate of 30% was made by IDC in 2016 [8]. For Indonesia, do observe nearby ads suggesting users upgrade. Lastly, users were the GSMA reports smartphone penetration of 47% [14]. While both also keenly aware of performance expectancy, “the degree to which estimates are not extremely different from our combined 3G and4G an individual believes that using the system will help him or her to observations, neither source disaggregates smartphone adoption attain gains in job performance,” where job performance would be across rural and urban boundaries. observed user activity like watching movies or learning new skills. In this section, we describe the two communities of study in this There is also work more specifically focused on mobile phone project: Barangay San Andres in Luzon, Philippines, and Bokondini adoption. Van Biljon et al. [32] provide a thorough review. The in Papua, Indonesia. It’s worth noting that the study is not compar- authors create a framework for viewing mobile phone adoption ing the two sites against each other (as their differences are many) through either physical, social, mental, or technological contexts. and it is instead seeking findings that generalizes across both (as Our work focuses on the technological (2G vs. 4G), social (media their commonalities include having community cellular networks drives for smartphone adoption) and physical (rural areas) contexts. and being rural). An Investigation of Phone Upgrades in Remote Community Cellular Networks ACM ICTD ’17, November 2017, Lahore, Pakistan

Figure 2: The two cellular networks, San Andres (left), Bokondini (right)

the tourism activities in the Barangay through further development of tourist attractions and bolstering of the local tourism industry. 3.1.2 Infrastructure. San Andres does not have access to an elec- tricity grid, but instead has a communal generator that operates for Figure 1: The Location of Barangay San Andres a limited duration each day. Billing for this service is organized in an ad-hoc manner with a member of the local government canvassing the community at night to see who is using power. Some house- holds have their own generator or use solar power. Deep wells, 3.1 Barangay San Andres aside from the small streams, are the main source of water. There Barangay San Andres is a rural Barangay (a barangay is the basic is no Internet access available in the community and instead users political unit in the Philippines) within Tanay, which is a first class travel outside to make use of free or inexpensive wifi elsewhere. municipality in Rizal province (Figure 1). Barangay San Andres is 3.1.3 Cellular Network. A community cellular network using located within the watershed of the Upper Marikina River Basin the CommunityCellularManager [11] was installed in San Protected Landscape and has a generally mountainous terrain. It is Andres in February 2016 (see Figure 2). The equipment supports relatively near (approximately 60 km away) Quezon City, which is voice and SMS only and is not yet Internet-capable. The base station a highly urbanized city that was once designated as the nation’s is solar-powered to compensate for the variance in local power gen- capital. It is also within easy commuting distance (around 30 km eration. Subscribers are required to purchase a local SIM (branded away) to Tanay Town proper. Tricycles (three-wheeled motorcy- TM Barangay) to connect to the local cellular network. The pres- cles) are the main mode of public transportation in and out of the ence of cellular signal has changed the way locals do business. The Barangay. tourism industry got a boost, as tour guides are now capable of The total population of Barangay San Andres is 2,145, with 481 contacting clients and maintaining a Facebook page with their local households as of December 2016. It is composed of 1,122 males and cell numbers. The network was installed in partnership with Globe 1,023 females. There are approximately 228 students studying at the Telecom and uses their satellite backhaul. elementary school located within the Barangay, while 19 students are enrolled at the kindergarten within the Barangay. 3.2 Bokondini, Papua, Indonesia 3.1.1 Industry. Barangay San Andres is largely rural, with stre- Bokondini (Figure 3), our second data collection site, is a small tches of farmlands, grazing areas for livestock, mountainous areas, rural community situated in the highlands of Papua, a mountainous and geologic formations that attract tourists and sports enthusi- region traversing the center of the island of Papua in Southeast asts. As such, the sources of income include farming, tourism, and Asia. It is approximately four hours drive from the regional capital the seasonal production of woven split bamboo mats (sawali). San of Wamena, which is only accessible by flying from other parts Andres features tracts of privately-owned farms where locals are of Papua. The road to Bokondini is paved, though it involves one employed as farmers and caretakers. Major cash crops include rice, serious river crossing that requires a 4-wheel drive truck. cassava, banana, and papaya. Bokondini is an established community of approximately 1500 Potential tourist attractions abound in San Andres, such as the people who are primarily indigenous Papuans. Unfortunately, there Sangab Cave, Kay-Ibon Falls, and Kanabuan Falls. The Barangay is no official census of the region, leaving only informal measure- also serves as the jump-off point for Mt. Batalusong. Hikers are ments. The community itself consists of five major groups: First, required to secure local tour guides from amongst San Andres. there are three local churches (all Protestant) each with their own Recent efforts of the Tanay Local Government aim to institutionalize local buildings. These churches are run primarily by, and service, ACM ICTD ’17, November 2017, Lahore, Pakistan K. Shah et al. the local Papuan Christian community. Outside of these churches, Table 1: Fields derived from TAC Data there are two local subcommunities for non-Papuan Indonesians (Pendatang), the police and the military. The police handle local Field Example community matters and the military maintain a small base to limit Marketing Name iPhone A1203 local participation in the Free Papua movement. Manufacturer Apple Inc 3.2.1 Industry. The region’s main economic activity is subsis- Radios GSM 1800, GSM 1900, GSM 900, GSM850 tence agriculture. One particularly notable cash crop is pineapples, Model Name iPhone A1203 which are known to be good in Bokondini. In the morning, vendors Operating System iOS bring pineapples by taxi to major markets in Wamena. Apart from Device Type Handheld agriculture, there are a few stores and a large amount of political Radios Bluetooth 2.0 and governmental activity as the village transitions into a regional Manufactured June 2007 district.

3.2.2 Infrastructure. Bokondini has no community infrastruc- 4 DATA AND METHODS ture, including no traditional cellular coverage. Local organizations The primary data analyzed in this work is logon (“camp” in cellular and individuals operate decentralized water, power, and Internet ac- terminology) records from two rural community cellular networks. cess solutions. There is an airstrip which is used by the Missionary We conduct a series of quantitative analyses of this dataset to learn Air Force for emergencies and occasional resupplies. There is also more about phone adoption behavior, including specific technolo- a road running to Wamena, which is the closest major city with gies and rates of adoption, across the two communities. Following existing power, water, and cellular network access. these results we drive deeper by conducting a series of qualitative interviews inside of one of the communities to explore why and 3.2.3 Cellular Network. In 2013, as part of early work on com- how people are adopting new devices. munity cellular networks, a local cellular network was installed in Bokondini [17]. The network is operated by a local ISP in Wa- 4.1 Quantitative Data and Analysis mena and managed by a primary school within the community The principle item of data being analyzed in this work is the Type (which is summarily owned and operated by one of the local Papuan Allocation Code or TAC of the mobile phone subscribers. The TAC churches). The network tower, a tree, is shown in Figure 2. The is a subset of the International Mobile Equipment Identity (IMEI), network has profitably operated for the following three years with- which is akin to the serial number of a cellular phone. An IMEI is a out any major hiccups. In late 2016, the network operator decided fifteen digit code formatted as follows: to change their pricing to a monthly base fee with usage on top, AA − BBBBBB − CCCCCC − D which caused a large culling of underused accounts. This dropped the number of subscribers in the system. This network, like the Where the A and B sections specify the serial number of that specific Philippines network, is a 2G voice and SMS only network utilizing phone, the C block specifies the TAC and the D digit is a checksum. the CommunityCellularManager software. Data remains unavail- From an IMEI, we can indentify the unique phone (using the serial able to subscribers. It is backhauled over a a satellite terminal to number) and the model of phone used (using the TAC). Hong Kong and uses phone numbers from the online provider To track upgrades of phones we also collected the International Nexmo. Mobile Subscriber Identity (IMSI) from all phones that attempt to attach to the towers. The IMSI uniquely identifies a SIM card, and when paired with the IMEI, provides both unique identifiers for a SIM and phone pair. When a single device (IMEI) is observed with multiple SIM cards (IMSI), we call this a “shared” phone, but note that this is likely an underestimate of actual phone sharing, since we cannot tell when two people share a SIM card. From the TAC data, we cross-reference with online databases to infer characteristics of the mobile phones themselves. The key fields, shown in Table 1, make it possible to understand the features and affordances of the phones in our network. We also sourced the manufacturing date of the devices by looking up each on the Internet (usually GSMArena [15]). We were able to find the man- ufacturing date of most 3G and 4G phones, but many 2G phones were not available. Outside of these datasets, no other personally identifiable in- formation was utilized, such as participant names or social media accounts. 4.1.1 Data Collection. Our data was collected over two and a Figure 3: The Location of Bokondini, Papua half years in Bokondini (August 2014 to April 2017) and one year in An Investigation of Phone Upgrades in Remote Community Cellular Networks ACM ICTD ’17, November 2017, Lahore, Pakistan

San Andres (April 2016 to April 2017). As part of the GSM protocol, longitudinal view we have of the network, we can remove the case a phone attempts to camp on a tower by sending a Location Update where the phone is returned to its original owner. Request (LUR) to the tower. This LUR contains both the IMSI and Definition: A user has upgraded their phone IMEI of the phone attempting to connect. It is from these connection when they change their primary phone to another attempts that we gather the quantitative data used in the study. and do not switch back. The GSM protocol also specifies how phones select their tow- ers. A phone first scans the supported frequency bands for towers Phone Sharing. Analogous to phone upgrades, we can also quan- available. From that list of towers it selects ones that match the tify the amount of phone sharing in the network. In this case, provider’s SIM card. If none of those are available, it will attempt instead of a user changing from one phone to another, we count to connect to whatever other towers are present. Since the towers how often a phone changes to another user and back. While not we observe are the only active towers in the region, this means all directly related to phone upgrade behavior in the network, this phones in the area attempt to camp. As such, our TAC records are statistic will provide a deeper understanding of user behavior in comprised of all phones that pass through the coverage area these networks and potentially inform mechanisms for increasing of the two access points and provide a complete, ground truth 4G adoption. view into the phones being used in this area. Definition: A phone is shared if it changes users with operation returning to earlier users. 4.1.2 Operational Definitions. Local vs. Non-local Subscribers. In subsequent analyses we will 4.2 Qualitative Interview Design differentiate between local and non-local users. We define local To get a better understanding of the 4G ‘mobile phone frenzy’ [23] in users as those who either bought a SIM on the community cellular the Philippines, we conducted semi-structured, topical interviews in network, or had their SIM active on the local network for at least San Andres. Unfortunately we could not do the same in Bokondini 10 days. This latter criteria was applied to deal with the fact that due to political instability in the area during the study. San Andres has a significant tourism industry, and we don’t want Phenomenology is the qualitative research paradigm that in- these transient phones to skew our results toward a more urban set formed the conduct of the topical interviews on respondents’ expe- of user demographics. 1 rience of using and upgrading their mobile phones. The focus is on collecting descriptions of mobile phone users’ lived experience to Definition: A local user is one who has a SIM for come up with a composite description and an interpretation of this the local telecom or has had their phone present experience [9]. Because the interviews were one-sitting interviews, in the community for at least ten days. the researchers tested for communicative validity in analyzing the Non-local phones are those who camp on the network but do empirical material. Thus, the qualitative responses of the interview not meet the above criteria. participants were compared across several demographic criteria, Primary Phone. While it’s easiest to assume a user doesn’t change such as age, gender, occupation, and nominal descriptions of level phones often, prior work has shown that sharing of phones and of technological background, to identify the most coherent inter- SIMs is common in much of the developing world [3, 10]. As such, pretation that would explain their experiences as a collectivity [28]. attempting to determine if a user has upgraded their phone could be confusing. To mitigate this concern, we introduce a notion of a 4.3 Data Collection Limitations primary phone. Unfortunately, the data we were able to collect in this study was uneven. For instance, we gathered hourly TAC logs for the Philip- Definition: A user’s primary phone is the phone pines but not for Indonesia. This is because the two networks are their SIM registered to the network with most of- separate and we did not set one policy for both. We gathered two ten during a day. and a half years of logs for Indonesia but only one year for the For example, a participant may use an iPhone 3G as their primary Philippines. This is because the Filipino network was installed later. phone for 2 months, while perhaps letting others install their SIMs We conducted interviews in the Philippines and not Indonesia, be- in the phone for a few minutes a day. cause of political issues in Bokondini. This lack of symmetric data Phone Upgrade. As one of the main questions we seek to answer collection limits the breadth of the study. with this data is why and how users are upgrading to 4G, we need to define what we consider to be an upgrade. As mentioned above, 5 RESULTS users often share both SIMs and phones, so we cannot naively 5.1 Local vs Non-Local Users count any change in phone as an upgrade. Instead, because of the We begin by analyzing the differences in phone types for local and non-local users in both communities. As can be seen in Figure 4, 1For example, 4000 IMSIs (out of 10000 total) were present for only one day in the local and non-local users in Indonesia are quite similar in their total lifetime of the Indonesia network. Similarly, for the Philippines, 70000 IMSIs (out of 100000 total) were present for just one day. To determine which of the population use of feature phones vs. smartphones: local phones were 60.0% should be considered local to the area, we plotted the number of SIMs active over 2G and 11.2% 4G; non-local phones were 58.8% 2G and 15.2% were number of days in the network. The plots are monotonically decreasing and we use the 4G. By contrast, differences between local and non-local devices in elbow point in the graph of 10 days to filter local users from non-local users. Filtering users with this estimate gives us 968 unique IMSIs in Bokondini and 2600 unique IMSIs the Philippines were pronounced. There, 51% of local phones are in San Andres. 2G while 15.9% were 4G; for non-local phones, only 27.4% were 2G ACM ICTD ’17, November 2017, Lahore, Pakistan K. Shah et al.

Figure 4: Local and non-local phones in use in past year in Figure 5: Distribution of local phone types over time in In- Indonesia (top) and Philippines (bottom). There is little dif- donesia (top) and Philippines (bottom). 2G phones are red, ference for local and nonlocal in Indonesia, but a large dif- 3G green and 4G blue. 3G growth has stagnated and stopped ference for Philippines. in both communities while 4G has grown significantly.

while 37% were 4G. For this reason, unless stated otherwise, the analyses that follow focus on local subscribers of both networks.

5.2 Phone Adoption We next analyze trends in smartphone adoption over time in the two rural networks. The results are shown in Figure 5. In Bokondini, where we have nearly three years of data, the changes are dramatic. Between August 2014 and April 2017, 4G adoption increased from 0.2% to 17.9% (a 90x increase) in the last month. In the last year alone, adoption rose from 6.5% to 17.9%, a growth of nearly 2.75x. This was primarily done at the expense of feature phones, as the Figure 6: Projected local 4G growth in Indonesia 3G adoption rate remained relatively unchanged throughout the entire trial period (22.5% in October 2014 to 25.9% in April 2017). may soon be able to leverage this trend to install 4G-only basesta- Feature phone (2G) adoption dropped from 77.3% to 56.3% over the tions even in these most remote areas. Moreover, these upgrades same period of time. are seemingly skipping 3G, with users replacing their 2G feature For San Andres, the growth in 4G adoption was strong, though phones with more modern 4G smartphones. not as pronounced as in Indonesia. In the year of data we observe, In Figure 6, we show the projected growth of 4G adoption in adoption rose from 11.8% in April 2016 to 22.5% in April 2017 (2x Bokondini from now until 2019, based on a model fit on the data year over year). 3G adoption again remained constant, with 35.3% collected between 2014 and 2017. Specifically, we fit a quadratic adoption in 2016 to 33.6% in 2017. 2G feature phone dropped from regression of number of monthly users Y on time t, as Y = α + 53% to 43.9%, and is no longer a majority of the total phones in the t t β t +β t2 +ϵ . We show only Indonesia because the relatively short area. 1 2 t period of data collection in the Philippines (1 year) is insufficient for robust projections. 5.3 Projecting Viability Bokondini had roughly 100 active 4G phones in the area as of The above results tell a compelling story about technology shift in April 2017, and the projections indicate that there will 200 active rural areas. 4G smartphone adoption is growing rapidly in both net- phones by August 2018, and 300 active 4G phones by March 2019. works despite the lack of general Internet connectivity. Operators From informal discussions with the local carrier, they estimate that An Investigation of Phone Upgrades in Remote Community Cellular Networks ACM ICTD ’17, November 2017, Lahore, Pakistan

Table 2: Local Phone Upgrades Table 3: Age-Statistics of “New” Phones in Philippines

From: To: Feature phone To: 3G To: 4G Mean Median Std. Dev Num. Phones Panel A: Philippines 2G to 3G 2.53 years 2 years 2.11 years 267 Feature phone 39.75% 3.56% 13.79% 2G to 4G 1.89 years 2 years 1.16 years 93 3G 4.70% 3.58% 3.18% 3G to 4G 1.65 years 2 years 1.19 years 87 4G 12.38% 3.66% 15.51% 4G to 4G 1.38 years 1 year 1.05 years 75 Panel B: Indonesia Feature 55.38% 1.32% 13.26% Table 4: Age-Statistics of “New” Phones in Indonesia 3G 1.60% 1.65% 0.85% 4G 14.73% 0.94% 10.25% Mean Median Std. Dev Num. Phones 2G to 3G 2.73 years 2 years 1.67 years 67 this target of 300 4G phones present in the addressable market 2G to 4G 0.66 years 1 year 0.49 years 12 is the break-even level at which point it is financially sustainable 3G to 4G 0.85 years 1 year 0.39 years 17 2 to deploy 4G technology. Assuming that people with 4G phones 4G to 4G 0.75 years 1 year 0.58 years 28 would use data services at an ARPU similar to the existing 2G services (which remains to be seen), these back-of-the-envelope projections indicate that the network would likely be financially with users upgrading to phones that are, on average, over a year old. viable in early 2019. In Indonesia, users are instead upgrading to more modern phones 5.4 Phone Upgrades that were produced within the last year. In Indonesia, there is no clear difference between the upgrade ages given the user’s previous As discussed in Section 4.1.2, we define a phone upgrade as when a device; a 2G phone holder may upgrade to a more recent device user moves from one primary phone to another without reverting than a 3G phone holder. In the Philippines, users of a 4G phone are to their original device. Table 2 shows the upgrade matrix for both more likely to upgrade to a recently manufactured phone than 2G San Andres and Bokondini, where a row represents the old device or 3G users. This may be a wealth effect, or just a measure ofthe and the column represents the new device. For instance, 39.75% upgrade cycles present in the Philippine secondary market. of all upgrades in the Philippines are from one feature phone to another. The results broadly support the phone adoption rate seen 5.6 Phone Sharing in the prior section; users are skipping 3G entirely and only moving between 2G feature phones and 4G smartphones. Since the starting Phone sharing is a common practice in many developing coun- population of 2G phones is already very high, the 14% of feature tries [5, 6, 21], and in many ways is a counterpoint to phone up- phone upgrades going to 4G represents a significant fraction of the grading. While we cannot distinguish between two individuals who entire network. share the same device and SIM card, we can detect when a single It is also worth noting that it appears there was nearly equivalent device (IMEI) is used by multiple SIMs. Figure 7 shows, for each amounts of phone downgrading as upgrading for 4G users. This is type of device, the fraction of those devices that are used by more not something we expected, and to our knowledge has not been than one SIM. On running a parametric t-test, we observe that 2G phones are shared significantly more than 3G and 4G phones in previously documented. This downgrading behavior may indicate −16 that users are shifting to 2G phones due to the lack of a 4G network. both the communities (p < 10 in both cases). That implies there may be latent 4G demand that could manifest The fact that feature phones are more frequently shared than if the network were upgraded. It also suggests that 4G phones are smartphones is potentially counterintuitive; as a scarcer commodity, not appropriate for these rural areas; in our interviews we discover it would likely be in higher demand, and in some cases people might that they consume too much power and are too fragile to survive pool resources in order to upgrade. On the other hand, it could be in difficult conditions (see Section 5.8). that smartphones are more expensive, more fragile, have shorter battery lives, or other factors that might discourage sharing. Here, 5.5 Upgrade Age our qualitative interviews provide some insight: many subjects had negative experiences with losing or damaging their primary We also investigate the age of phones being upgraded. For this, we phones, often from sharing. Another factor that participants cited use the TAC data (Section 4.1) to query an Internet database that was the relative ease of sharing 2G phones: they are simple to use, tells us, for most phones, the manufacture date of the phone. This quick to set up, and utilize less unique information about the user. is compared against the time the upgraded phone was first present on the network. As can be seen using Table 3 and Table 4, the San 5.7 Phone Activity Andres appears to have a much more robust secondary market than Bokondini (perhaps due to the proximity to a major urban area), Our final set of quantitative results explore differences in the extent to which feature phones and smartphones are active on the San 2Specifically, they report the need for 120 active subscribers, with roughly 40% pene- tration into the local subscriber base. This means that we wish to know when there Andres network. Figure 8 shows, by device type, the average num- will be 300 total 4G phones local to the Indonesia network. ber of hours per day for which the device is active on the network, ACM ICTD ’17, November 2017, Lahore, Pakistan K. Shah et al.

Figure 7: The local phone sharing ratio in Indonesia (left) and the Philippines (right). Features phones are red, 3G green and 4G blue. 2G phones are shared much more than 3G or 4G phones.

without Internet connectivity, we conducted interviews in San Andres, following the protocol described in Section 4.2. We were able to interview two of the five SIM card dealers in San Andres; the other three SIM card dealers were not available. In separate interviews, we asked the dealers to give us a general description of their clients (buyers of SIM cards and/or pre-paid airtime). We also asked them about their general observation about their clients’ cellphones. The two SIM card dealers note that most of their clients are young people and that younger people in the area own "touchscreen phones" (i.e., smartphones) while the older ones own "keypad phones" (i.e., feature phones). This information Figure 8: Hours of local Filipino network activity over time. became the basis for initiating the ’snowball’ technique in identify- 2G phones are significantly more active than smartphones. ing respondents. We positioned ourselves in public places in the Barangay and approached young adults for interviews. After each interview, the researcher asked the respondent to refer additional as constructed from the hourly TAC logs.3 The results show that respondents in the Barangay including possible respondents who users with 2G phones keep them active more than users with 3G are no longer young adults. In total, we conducted 15 topical inter- phones, who again are more active than users with 4G phones. This views with individual users, in addition to the two interviews with is again somewhat counterintuitive; if you have a 4G phone for SIM card dealers. All interviews were conducted in San Andres in media purposes, wouldn’t you use it more? One likely hypothesis Tagalog by native speakers. for keeping your 4G phone inactive is due to the lack of community The age range of interview participants is from 18 to 67 years of power infrastructure in the area. Despite the better media playing age, where nine (9) are within the age range of 18 to 30 or young properties of smartphones, they need to be turned off to conserve adults and six (6) are older than 30 years old. Ten of the fifteen power in areas where you aren’t leveraging the other 4G capabili- participants are female. Respondents’ primary occupation is as ties of the phone. Secondly, 2G phones are simply more efficient in homemaker (5), otherwise employed (4), students (3), unemployed idle mode; some generic 2G feature phones can operate with over (2), and retired (1). The researchers classified the interview partici- three days of standby time, significantly more than a power-hungry pants according to the latter’s mobile phones during the interview smartphone. as follows: 2G = 4 participants, 3G = 4 participants, 4G = 7 partici- pants. 5.8 Philippines Interviews All of the interview participants reported that their exposure to The prior figures and tables highlight several prominent trends ICTs is mainly through the use of mobile phones. All of them are and regularities of phone use in two rural communities. To provide aware that there are mobile phones that can do much more than more nuance and depth to these results, and in particular to develop facilitate communication, although young adults are better able to a better understanding of the role of smartphones in communities describe the other capabilities of their mobile device) [27]. Interview respondents do not refer to their mobile phones as 2G, 3G, or 4G, but 3This registration implies that the user’s phone was active during that hour. As such, young adults are aware that 3G and 4G phones are more powerful a phone that sends more registrations in a day was actively powered-on more than than 2G. Interview respondents classify phones by describing the one that did not. There are some caveats; phones in will not attempt to register and a phone that is on for one minute will be counted as often as a phone that phone’s capabilities, thus phones are either touchscreen or keypad was powered on for fifty-nine minutes. Unfortunately, we gathered only daily TAC and Internet-capable or not. As a 22-year old female respondent logs in Bokondini and do not have the granularity to measure phone usage per day who owns a 4G mobile phone describes it: for that network, as we only know if a phone was active at all that day and not their amounts of activity relative to other phones. An Investigation of Phone Upgrades in Remote Community Cellular Networks ACM ICTD ’17, November 2017, Lahore, Pakistan

Respondent 7:"If you have a touchscreen phone, you can do many things with it. You can access the internet. Unlike before, you can only do calls and text. But now, (with a touchscreen phone), you can call, text, (connect to the) internet, and other things." Except for two (the two oldest), almost all interview participants have upgraded their phone models because of their desire for a more powerful phone. Those two older interview participants have actually downgraded their phone type from a smartphone to a feature phone because they found feature phones more durable and sufficient for their needs (e.g. calls and ). Figure 9: Examples of Filipino advertisements, with media General observations. Interview participants use their phones downloads (left) and youth targeting (right). as communication devices. The nature of their communication is predominantly personal, involving family members, friends, and loved ones, especially those who live and/or work outside of San quantitative work: many individuals report using feature phones Andres. Almost all of the interview participants have a Facebook within San Andres, but then switching to a smartphone when out- (Facebook Lite) account to maintain existing kinship ties, to re- side of the Barangay. This behavior, if anything, would lead us to establish old personal ties, and to build new ties. Communication underestimate the demand for 4G services, as seen in Section 5.3. may also involve employers or business associates. During the interviews, verbal and non-verbal cues indicate that Entertainment. Inside the Barangay (and outside of Internet con- young adults seem dissatisfied and unimpressed with the local com- nectivity), phones are commonly used as a cheaper entertainment munity cellular network because it could only provide call and text device. In addition to the camera - which was one of the most im- messaging services. While the Philippines is commonly referred to portant functions to many participants - commonly reported uses as "the texting capital of the world" [24] we find that in the remote include listening to music, watching movies, and watching televi- rural Barangay of San Andres, people with 4G phones minimize sion shows (especially local, Korean, and Japanese shows). Through texting and voice calls because it is more expensive when compared their network of friends outside of San Andres or by occasionally to traveling to nearby town centers to access free WiFi services for getting connected to the Internet, they are able to store and share the public. They prefer to use instant messaging platforms such as entertainment files that could be re-shared or enjoyed together Facebook Messenger, Snapchat, and WhatsApp in these places be- with family and friends in San Andres, where they have very lim- cause communication is real-time. Two participants preferred to be ited entertainment options. Apps that are used for these activities interviewed together and their joint account was quite descriptive: include Share.It and Youtube. Cellphones are also commonly used Respondent 3:"We still don’t have internet here." as a gaming device. Games, like music and videos, are copied or Respondent 4:"Sometimes, (we get cellular signal) downloaded onto their cellphones. This is particularly true for those in the mountains. We go there to send a text message who are less mobile, who place a high premium on phones with to our boyfriends. better audio/video, faster performance, and that facilitate sharing Respondent 3:"Then, you can’t be sure if the mes- of content. As a 25 year-old tricycle driver who has a 3G phone sage was sent..." narrates: Respondent 4:"We don’t send text messages any- Respondent 12:"This cellphone belongs to my wife, more. We send PMs through Facebook Lite, Viber, but my child uses it. My child is seven years old. We WhatsApp... use it just for games, music, Facebook. (Facebook) Respondent 3:"Now, you can see if they are online. is the only (app) that we know (how to use) here. I You know immediately if they are replying. Unlike let him download as many games as possible, then I before when you need to wait..." will borrow the phone from him. Respondent 4:"You will wait like waiting for a It is worth noting the emergence of social networks of enter- telegram..." tainment enthusiasts that facilitate file-sharing. In pursuing this Thus, while San Andres does not have 4G connectivity, partici- analytical lead, we did a quick ocular visit and initial content anal- pants report traveling to places outside of the community with free ysis of materials in nearby commercial centers. These revealed that WiFi to make use of data. In a sense, they are able to transgress cellphone dealers and repair shops in malls have matched this user- the limitations of their physical and social context by traveling defined desire with commercial services such as downloading of to areas where the services they require are more accessible. The entertainment files for a fee (see Figure 9). It is equally interesting structure of the telecommunications market facilitates this process: to note that according to respondents, families share entertainment consumers in the Philippines commonly own multiple SIM cards files that are stored in the mobile phone of a family member. They from multiple operators, and frequently switch to the SIM card say that this is a more affordable alternative compared to going to that is most relevant and affordable to the local market in which the movies or to subscribing to cable television as sources of family they inhabit. This helps explain the dominance of 2G phones in our entertainment. ACM ICTD ’17, November 2017, Lahore, Pakistan K. Shah et al.

Upgrades and downgrades. To contextualize the upgrade behav- these rural areas, by their nature of being rural and low-income, ior that was a focus of earlier sections, we asked participants about lack exactly this kind of infrastructure. their reasons for changing phones. Women respondents say that they would have to immediately replace their cellphones when Driving Entertainment. One primary finding from the interviews these are lost, stolen, or permanently damaged because mobile was the importance of entertainment as a driver of 4G/smartphone phones are necessary for them to keep in touch with their loved adoption. Over half of the interviewees noted an entertainment ones especially those who live and work outside the Barangay. For function (music, games, or video) as one of the primary uses for most respondents, regardless of gender or occupation, upgrades are their phone and a reason for upgrading to a newer device. Users justified when their mobile phone is prone to stalling or hanging used their phones to download, copy, store, and play audio and or has a limited storage capacity because then the phone’s enter- video files for both their own personal use as well as for community tainment function is hampered. Unplanned upgrades were also or family use. For example, one of the interview subjects, a 22-year frequently reported, where lower end (but still functional) phones old construction worker in San Andres, watches video and listens are gifted to loved ones, for instance to a younger sibling, a relative, to music from a memory card, which he loads with content from or even to an elder adult. outside of the community. Even in an area without cellular coverage, In determining what device to upgrade to, affordability is criti- users batch store this content on their smartphones for distribution. cal. Higher end cellphones from cheaper brands proliferate in San There may be an opportunity to encourage 4G phone adoption Andres. These units are bought either as brand new phones or by providing media that can most efficiently be shown on 4G smart- second-hand (used). Aside from cost, phones are deemed better phones. Examples such as a Youtube Accelerator [13] allow local or worse in terms of functionality: faster, bigger storage capacity, caching of audio or video content and could provide support for wider screen, better camera, and better video graphics. Aesthetics locals to buy a 4G phone to use in the community. also matter, and slimmer models are considered more stylish. “Hardening” Devices. The primary reason for switching phones given in the interview was loss of the device; theft or destruc- 6 DISCUSSION tion. For example, the 22-year old construction worker discussed The smartphone frenzy in remote and isolated villages may seem above had upgraded from a feature phone to a cherry mobile (a irrational, given that these communities do not have 4G network local smartphone) phone. He then upgraded to a nicer smartphone infrastructure in place. Yet our quantitative and qualitative analysis after dropping the cherry mobile phone in some water. Another shows that smartphone use is quite prevalent and provides some way to encourage 4G adoption would be to harden smartphones insights to explain this phenomenon. Here, we summarize a few of against these types of loss. Some feature phones are notoriously re- these key insights, and discuss how they can inform the broader silient and having smartphones with the same ruggedness would be discussion around connectivity in developing regions. better-suited to the conditions of harsh rural environments where power and weather often destroy hardware [31]. Similarly, it may 6.1 Supporting 4G Adoption be that allowing for repair is the answer. Prior work has explored the incredible depth of phone repair in developing regions [20]. Infrastructure Support. One key finding of this work is that while Unfortunately, the social infrastructure and knowledge required there is still robust uptake of 2G phones, the market for 3G-only for repair is often unavailable in rural areas. phones is dying. This means that any attempt to increase smart- Theft is a trickier problem. For instance, a 28-year old female phone adoption will involve increasing adoption of phones that homemaker in San Andres was forced to downgrade from a Sam- support 4G. While operators may still wish to deploy 3G networks sung Galaxy to a Cherry Mobile phone when the Samsung was for other reasons (e.g., patent encumbrances on 4G or extra hard- loaned to a friend of her child and not returned. As part of a com- ware in a warehouse) there will be no reason to do so based on munity network, it is possible we could provide stronger tooling for adoption in the near future. Indeed, we note the phone downgrading detecting and thwarting theft of devices within our network. Major results from Section 5.4 that show that 4G users in our network telecoms have, in the past, provided a blacklist service for blocking switch back to 2G phones as much as they upgrade to 4G phones. stolen IMEIs, but that feature is not well supported. Implementing This implies that these users entered the rural area with a 4G phone a similar service within the community network could discourage and decided to switch back during their time in our networks. phone theft, but if the phone left our community we could no longer It seems likely that if there were 4G coverage available in the enforce any theft policy. community these users would be incentivized to stick with their existing 4g phone. The capabilities lost in the step down to a 2G Policy Rationale. There are many benefits of moving rural users network would mitigate many of the advantages of the device. Sim- to 4G handsets. Users gain access to a more feature-rich device ilarly, during our interviews users noted that power was a primary that can access both traditional telephony as well as Internet ser- limitation on their use of smartphones. This is supported by our vices. This device can help them engage with increasingly online measurements in Section 5.7, where we found that 4G phones were government and financial services. Lastly, as the interviews clearly less active on the network (in terms of registrations) than 2G or 3G noted, the smartphone is a much better device for consuming enter- phones. As such, the data (both TAC logs and interviews) seem to tainment. Similarly, it would be largely beneficial to operators and imply that better infrastructure (both network and power) in a com- regulators if the users transitioned to 4G. Refarming of 2G spectrum munity would increase 4G adoption. Solutions like microgrids [22] would create more bandwidth as well as allow for a sunsetting of or joint 2G/4G basestations may be what is needed. Unfortunately, legacy 2G cellular infrastructure, which is extremely expensive. An Investigation of Phone Upgrades in Remote Community Cellular Networks ACM ICTD ’17, November 2017, Lahore, Pakistan

One key thing about phone upgrades is that, from a user perspec- to upgraders because of the absence of necessary physical and so- tive, each network technology is a superset of all prior technologies. cial infrastructure that would enable them to use their cellphones This means that every 3G phone supports both 2G and 3G and beyond personal communication and personal entertainment. every 4G phone supports 2G, 3G, and 4G. This means that 2G bases- Where, then, is there room for change? In this study, we found tations will continue to be relevant even as 4G phones become it in the noticeable practices of small acts of transgression, or what the dominant paradigm. There is an option for operators to install James C. Scott referred to as ’weapons of the weak’ [29], to cir- dual 2G/4G basestations. While not a common model yet (due to cumvent context by cellphone users in a remote rural village. As the fact that 2G networks often already exist) these would be a previously mentioned, cellphone users change SIMs or travel to great mechanism for encouraging 4G adoption while supporting other places to take advantage of free wifi and avoid the costs of the current install base. texting and voice calls. This lived experience apparently leads to Of course, the big limiter on existing operator 4G rollouts is the feelings of dissatisfaction with poor (inaccessible and expensive) lack of backhaul in rural areas. 2G access points can be serviced connectivity services, and therefore experience with mobile phone by a (relatively) small satellite terminal (as they are in both of our use raises technological consciousness even if users have a low level community networks). A 4G base station usually requires a fiber of technological consciousness via other gadgets to begin with. connection, which is extremely expensive, and therefore rare, in rural areas. This means it may be the case that user adoption of 4G phones will outpace roll-outs of 4G networks. 6.3 Study Limitations Generalizability. This study is a qualitative and quantitative anal- Possible Downsides of 4G. While we believe the shift to ubiqui- ysis of phone upgrades in two specific rural community cellular tous 4G networks and mobile phones to be a net positive, there networks. These networks have a number of unique properties are negative elements to the shift away from 2G. In particular, that imply that the results may not generalize. The first is that as mentioned above, smartphones are power hungry and fragile. these are community cellular networks. Phone adoption in more Also, while they are getting cheaper, they are still more expensive traditional incumbent networks may be different. Those networks than a similar-year 2G phone. Despite potential solutions to these will likely support data service and better integrate with the rest problems, having 4G-only basestations would require users to buy of the country-wide system. Second, these networks sit in certain devices that may not fit their environment. It is our hope that, as4G rural areas with their own unique properties. Though the results networks become ubiquitous, extremely low-end 4G phones will have generalized across two very different communities in different start to be manufactured. These may even have the affordances of countries, there are likely to be changes in other communities. a traditional feature phone.

Future Networks. Though our focus is on LTE, technology com- 7 CONCLUSION panies are continuing to develop new wireless protocols that will We are in the midst of a massive shift from 2G cellular networks to eventually be present on future phones. It’s reasonable to ask if modern IP-based 4G networks. Caught in this shift are rural areas and when the next generation () networks will be relevant in that are currently uncovered or undercovered. In these communi- rural areas. As the first prototype LTE networks went live in nearly ties, operators are currently deploying aging 2G technologies as a decade ago in 2009 and the first 5G networks are just now be- there are not enough potential customers with devices that sup- ing deployed, one might expect a similar wait for their wide-area port 4G connections. At the same time, users are discouraged from adoption. As for now, there’s plenty of room for growth in LTE. upgrading as their local network does not support the many affor- dances of their 4G phones. 6.2 Contextualizing Connectivity This work explored mobile phone upgrade behavior in two re- The trends we observe become more intelligible if we look beyond mote communities in the Philippines and Indonesia. We utilized the the lens of personal choice or preference. Weaving the broader TAC logs to provide a detailed record of all cellular phones in these social context into the interpretation of individual accounts, one two regions. We also conducted fifteen semi-structured interviews is able to render visible other powerful forces that are driving 4G in the Philippines to get a deeper understanding of user behavior. growth. For instance, the dominant marketing strategy of cellphone From these analyses we discovered that 4G phones are common manufacturers and telecoms is important. Even a cursory look despite the lack of 4G coverage, 4G adoption is accelerating, users at advertisements and other promotional materials (see Figure 9) are skipping 3G, 2G phones are more active and shared, and that in selling 4G phones reveal a marketing strategy that is mainly media was a primary driver of smartphone adoption. We conclude focused on consumerist constructions of a "youth lifestyle", which by suggesting that telecoms consider ignoring 3G rollouts and pro- is similar to what was earlier suggested by Carroll et al. [7]. While vide several potential mechanisms for accelerating 4G adoption in we might expect that sophisticated gadgets would target "higher- these rural areas. end" consumers, in practice the marketing strategy deliberately focuses on lower-end markets, even users in remote, rural villages. ACKNOWLEDGEMENTS Therefore, we are observing a systematic and rapid integration of remote rural villages into the global market and within the frame- We wish to thank Jenn Webster for assistance in writing.This work work of a global consumerist culture through ICTs. Yet, all of the was funded by Facebook and a grant from the Financial Services potential benefits from ICT adoption do not automatically accrue for the Poor program at the Bill and Melinda Gates Foundation. ACM ICTD ’17, November 2017, Lahore, Pakistan K. Shah et al.

REFERENCES University Press Manila. vii, 197 p. : pages. [1] Sohaib Ahmad, Abdul Lateef Haamid, Zafar Ayyub Qazi, Zhenyu Zhou, [25] Agha Ali Raza, Farhan Ul Haq, Zain Tariq, Mansoor Pervaiz, Samia Razaq, Umar Theophilus Benson, and Ihsan Ayyub Qazi. 2016. A View from the Other Side: Saif, and Roni Rosenfeld. 2013. Job Opportunities Through Entertainment: Virally Understanding Mobile Phone Characteristics in the Developing World. In Pro- Spread Speech-based Services for Low-literate Users. In Proceedings of the SIGCHI ceedings of the 2016 Internet Measurement Conference (IMC ’16). ACM, New York, Conference on Human Factors in Computing Systems (CHI ’13). ACM, New York, NY, USA, 319–325. https://doi.org/10.1145/2987443.2987470 NY, USA, 2803–2812. https://doi.org/10.1145/2470654.2481389 [2] Joshua Blumenstock, Gabriel Cadamuro, and Robert On. 2015. Predicting poverty [26] Rhizomatica. 2017. Rhizomatica. https://www.rhizomatica.org/. (2017). Accessed: and wealth from mobile phone metadata. Science 350, 6264 (27 Nov. 2015), 1073– 2017-05-7. 1076. https://doi.org/10.1126/science.aac4420 [27] Everett M. Rogers. 2003. Diffusion of innovations (5th ed.). Free Press, New York, [3] Joshua Blumenstock and Nathan Eagle. 2010. Mobile Divides: Gender, Socioe- NY [u.a.]. 576 pages. conomic Status, and Mobile Phone Use in Rwanda. In Proceedings of the 4th [28] Jörgen Sandberg. 2005. How do we justify knowledge produced within interpre- ACM/IEEE International Conference on Information and Communication Technolo- tive approaches? Organizational research methods 8, 1 (2005), 41–68. gies and Development (ICTD ’10). ACM, New York, NY, USA, Article 6, 10 pages. [29] James C. Scott. 1985. Weapons of the weak : everyday forms of peasant resistance. https://doi.org/10.1145/2369220.2369225 Yale University Press, New Haven. [4] Joshua E. Blumenstock, Michael Callen, Tarek Ghani, and Lucas Koepke. 2015. [30] Thomas N. Smyth, Satish Kumar, Indrani Medhi, and Kentaro Toyama. 2010. Promises and Pitfalls of Mobile Money in Afghanistan: Evidence from a Ran- Where There’s a Will There’s a Way: Mobile Media Sharing in Urban India. In domized Control Trial. In Proceedings of the Seventh International Conference on Proceedings of the SIGCHI Conference on Human Factors in Computing Systems Information and Communication Technologies and Development (ICTD ’15). ACM, (CHI ’10). ACM, New York, NY, USA, 753–762. https://doi.org/10.1145/1753326. New York, NY, USA, Article 15, 10 pages. https://doi.org/10.1145/2737856.2738031 1753436 [5] Joshua E Blumenstock and Nathan Eagle. 2012. Divided We Call: Disparities [31] Sonesh Surana, Rabin Patra, Sergiu Nedevschi, Manuel Ramos, Lakshmi- in Access and Use of Mobile Phones in Rwanda. Information Technology and narayanan Subramanian, Yahel Ben-David, and Eric Brewer. 2008. Beyond International Development 8, 2 (2012), 1–16. Pilots: Keeping Rural Wireless Networks Alive. In Proceedings of the 5th USENIX [6] Jenna Burrell. 2010. Evaluating Shared Access: social equality and the circulation Symposium on Networked Systems Design and Implementation (NSDI’08). USENIX of mobile phones in rural Uganda. Journal of Computer-Mediated Communication Association, Berkeley, CA, USA, 119–132. http://dl.acm.org/citation.cfm?id= 15, 2 (2010), 230–250. 1387589.1387598 [7] Jennie Carroll, Steve Howard, Jane Peck, and John Murphy. 2002. A field study of [32] Judy van Biljon and Paula Kotzé. 2007. Modelling the Factors That Influence perceptions and use of mobile telephones by 16 to 22 year olds. JITTA : Journal Mobile Phone Adoption. In Proceedings of the 2007 Annual Research Conference of Information Technology Theory and Application 4 (2002), 49–61. Issue 2. of the South African Institute of Computer Scientists and Information Technologists [8] International Data Corporation. 2016. Asia/Pacific Quarterly Mobile Phone on IT Research in Developing Countries (SAICSIT ’07). ACM, New York, NY, USA, Tracker. http://www.gsma.com/globalmobiletrends/. (2016). Accessed: 2017-05- 152–161. https://doi.org/10.1145/1292491.1292509 7. [33] Viswanath Venkatesh, Michael G. Morris, Gordon B. Davis, and Fred D. Davis. [9] John W. Creswell, William E. Hanson, Vicki L. Clark Plano, and Alejan- 2003. User Acceptance of Information Technology: Toward a Unified View. MIS dro Morales. 2007. Qualitative research designs: Selection and implementa- Quarterly 27, 3 (2003), 425–478. http://www.jstor.org/stable/30036540 tion. Counseling Psychologist 35, 2 (3 2007), 236–264. https://doi.org/10.1177/ [34] Mariya Zheleva, Arghyadip Paul, David L. Johnson, and Elizabeth Belding. 2013. 0011000006287390 Kwiizya: Local Cellular Network Services in Remote Areas. In Proceeding of [10] J. Donner. 2015. After Access: Inclusion, Development, and a More Mobile Internet. the 11th Annual International Conference on Mobile Systems, Applications, and MIT Press. https://books.google.com/books?id=BAlkrgEACAAJ Services (MobiSys ’13). ACM, New York, NY, USA, 417–430. https://doi.org/10. [11] Facebook. 2017. CommunityCellularManager. https://github.com/ 1145/2462456.2464458 facebookincubator/CommunityCellularManager. (2017). Accessed: 2017-05-7. [12] Vanessa Frias-Martinez, Jesus Virseda, and Enrique Frias-Martinez. 2012. Socio- Economic Levels and Human Mobility. Jourhnal of Information Technology for Development (Feb. 2012), 1–16. [13] Google. 2017. Youtube Accelerator. https://accelerator.withgoogle.com/. (2017). Accessed: 2017-05-7. [14] GSMA. 2016. Global Mobile Trends. http://www.gsma.com/globalmobiletrends/. (2016). Accessed: 2017-05-7. [15] GSMArena. 2017. GSMArena. http://www.gsmarena.com/. (2017). Accessed: 2017-05-7. [16] Kurtis Heimerl and Eric Brewer. 2010. The Village Base Station. In Proceedings of the 4th ACM Workshop on Networked Systems for Developing Regions (NSDR ’10). ACM, New York, NY, USA, Article 14, 2 pages. https://doi.org/10.1145/1836001. 1836015 [17] Kurtis Heimerl, Shaddi Hasan, Kashif Ali, Eric Brewer, and Tapan Parikh. 2013. Local, Sustainable, Small-scale Cellular Networks. In Proceedings of the Sixth International Conference on Information and Communication Technologies and Development: Full Papers - Volume 1 (ICTD ’13). ACM, New York, NY, USA, 2–12. https://doi.org/10.1145/2516604.2516616 [18] Kurtis Heimerl, Anuvind Menon, Shaddi Hasan, Kashif Ali, Eric Brewer, and Tapan Parikh. 2015. Analysis of Smartphone Adoption and Usage in a Rural Com- munity Cellular Network. In Proceedings of the Seventh International Conference on Information and Communication Technologies and Development (ICTD ’15). ACM, New York, NY, USA, Article 40, 4 pages. https://doi.org/10.1145/2737856.2737880 [19] GSMA Intelligence. 2014. Country Overview: Philippines Growth Through In- novation. https://www.gsmaintelligence.com/research/?file=141201-philippines. pdf. (2014). Accessed: 2017-05-7. [20] Steven J. Jackson, Alex Pompe, and Gabriel Krieshok. 2012. Repair worlds: maintenance, repair, and ICT for development in rural Namibia. In CSCW. [21] Jeffrey James and Mila Versteeg. 2007. Mobile phones in Africa: howmuch do we really know? Social Indicators Research 84, 1 (Oct. 2007), 117. https: //doi.org/10.1007/s11205-006-9079-x [22] P. Achintya Madduri, Jason Poon, Javier Rosa, Matthew Podolsky, Eric Brewer, and Seth Sanders. 2014. A Scalable DC Microgrid Architecture for Rural Electri- fication in Emerging Regions. [23] BBC News. 2000. Mobile phone frenzy hits Cameroon. http://news.bbc.co.uk/2/ hi/africa/910121.stm. (2000). Accessed: 2017-05-7. [24] Raul Pertierra, Eduardo Ugarte, Alicia Pinggol, Joel Hernandez, and Nikos Lexis Dacanay. 2002. Txt-ing selves : cellphones and Philippine modernity. De La Salle